of 27
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
1/27
Consciousness as Integrated Information:a Provisional Manifesto
GIULIO TONONI
Department of Psychiatry, University of Wisconsin, Madison, Wisconsin
Abstract. The integrated information theory (IIT) starts
from phenomenology and makes use of thought experi-ments to claim that consciousness is integrated information.
Specifically: (i) the quantity of consciousness corresponds
to the amount of integrated information generated by a
complex of elements; (ii) the quality of experience is spec-
ified by the set of informational relationships generated
within that complex. Integrated information () is defined
as the amount of information generated by a complex of
elements, above and beyond the information generated by
its parts. Qualia space (Q) is a space where each axis
represents a possible state of the complex, each point is a
probability distribution of its states, and arrows between
points represent the informational relationships among its
elements generated by causal mechanisms (connections).
Together, the set of informational relationships within a
complex constitute a shape in Q that completely and univo-
cally specifies a particular experience. Several observations
concerning the neural substrate of consciousness fall natu-
rally into place within the IIT framework. Among them are
the association of consciousness with certain neural systems
rather than with others; the fact that neural processes un-
derlying consciousness can influence or be influenced by
neural processes that remain unconscious; the reduction of
consciousness during dreamless sleep and generalized sei-
zures; and the distinct role of different cortical architectures
in affecting the quality of experience. Equating conscious-
ness with integrated information carries several implications
for our view of nature.
INTRODUCTION
Everybody knows what consciousness is: it is what van-
ishes every night when we fall into dreamless sleep and
reappears when we wake up or when we dream. It is also all
we are and all we have: lose consciousness and, as far as
you are concerned, your own self and the entire world
dissolve into nothingness.
Yet almost everybody thinks that understanding con-
sciousness at the fundamental level is currently beyond the
reach of science. The best we can do, it is often argued, is
gather more and more facts about the neural correlates of
consciousnessthose aspects of brain function that change
when some aspects of consciousness changeand hope thatone day we will come up with an explanation. Others are
more pessimistic: we may learn all about the neural corre-
lates of consciousness and still not understand why certain
physical processes seem to generate experience while others
do not.
It is not that we do not know relevant facts about con-
sciousness. For example, we know that the widespread
destruction of the cerebral cortex leaves people permanently
unconscious (vegetative), whereas the complete removal of
the cerebellum, even richer in neurons, hardly affects con-
sciousness. We also know that neurons in the cerebral
cortex remain active throughout sleep, yet at certain times
during sleep consciousness fades, while at other times we
dream. Finally, we know that different parts of the cortex
influence different qualitative aspects of consciousness:
damage to certain parts of the cortex can impair the expe-
rience of color, whereas other lesions may interfere with the
perception of shapes. In fact, increasingly refined neurosci-
entific tools are uncovering increasingly precise aspects of
the neural correlates of consciousness (Koch, 2004). And
yet, when it comes to explaining why experience blossoms
in the cortex and not in the cerebellum, why certain stages
of sleep are experientially underprivileged, or why some
Received 20 August 2008; accepted 10 October 2008.
* To whom correspondence should be addressed. E-mail: gtononi@
wisc.edu
Abbreviations: , integrated information; IIT, integrated information
theory; MIP, minimum information partition.
Reference:Biol. Bull. 215: 216242. (December 2008) 2008 Marine Biological Laboratory
216
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
2/27
cortical areas endow our experience with colors and others
with sound, we are still at a loss.
Our lack of understanding is manifested most clearly
when scientists are asked questions about consciousness in
difficult cases. For example, is a person with akinetic
mutismawake with eyes open, but mute, immobile, and
nearly unresponsive conscious or not? How much con-
sciousness is there during sleepwalking or psychomotor
seizures? Are newborn babies conscious, and to what ex-
tent? Are animals conscious? If so, are some animals more
conscious than others? Can they feel pain? Does a bat feel
space the same way we do? Can bees experience colors, or
merely react to them? Can a conscious artifact be con-
structed with non-neural ingredients? I believe it is fair to
say that no consciousness expert, if there is such a job
description, can be confident about the correct answer to
such questions. This is a remarkable state of affairs. Just
consider comparable questions in physics: Do stars have
mass? Do atoms? How many different kinds of atoms and
elementary particles are there, and of what are they made?
Is energy conserved? And how can it be measured? Or
consider biology: What are species, and how do they
evolve? How are traits inherited? How do organisms de-
velop? How is energy produced from nutrients? How does
echolocation work in bats? How do bees distinguish among
colors? And so on. Obviously, we expect satisfactory an-
swers by any competent physicist and biologist.
Whats the matter with consciousness, then, and how
should we proceed? Early on, I came to the conclusion that
a genuine understanding of consciousness is possible only if
empirical studies are complemented by a theoretical analy-
sis. Indeed, neurobiological facts constitute both challeng-
ing paradoxes and precious clues to the enigma of con-
sciousness. This state of affairs is not unlike the one faced
by biologists when, knowing a great deal about similarities
and differences between species, fossil remains, and breed-
ing practices, they still lacked a theory of how evolution
might occur. What was needed, then as now, were not just
more facts, but a theoretical framework that could make
sense of them.
In what follows, I discuss the integrated information
theory of consciousness (IIT; Tononi, 2004)an attempt to
understand consciousness at the fundamental level. Topresent the theory, I first consider phenomenological
thought experiments indicating that subjective experience
has to do with the generation of integrated information.
Next, I consider how integrated information can be defined
mathematically. I then show how basic facts about con-
sciousness and the brain can be accounted for in terms of
integrated information. Finally, I discuss how the quality of
consciousness can be captured geometrically by the shape
of informational relationships within an abstract space
called qualia space. I conclude by examining some impli-
cations of the theory concerning the place of experience in
our view of the world.
A Phenomenological Analysis: Consciousness as
Integrated Information
The integrated information theory (IIT) of consciousnessclaims that, at the fundamental level, consciousness is inte-
grated information, and that its quality is given by the
informational relationships generated by a complex of ele-
ments (Tononi, 2004). These claims stem from realizing
that information and integration are the essential properties
of our own experience. This may not be immediately evi-
dent, perhaps because, being endowed with consciousness
most of the time, we tend to take its gifts for granted. To
regain some perspective, it is useful to resort to two thought
experiments, one involving a photodiode and the other a
digital camera.
Information: the photodiode thought experiment
Consider the following: You are facing a blank screen
that is alternately on and off, and you have been instructed
to say light when the screen turns on and dark when it
turns off. A photodiodea simple light-sensitive device
has also been placed in front of the screen. It contains a
sensor that responds to light with an increase in current and
a detector connected to the sensor that says light if the
current is above a certain threshold and dark otherwise.
The first problem of consciousness reduces to this: when
you distinguish between the screen being on or off, you
have the subjective experience of seeing light or dark. Thephotodiode can also distinguish between the screen being on
or off, but presumably it does not have a subjective expe-
rience of light and dark. What is the key difference between
you and the photodiode?
According to the IIT, the difference has to do with how
much information is generated when that distinction is
made. Information is classically defined as reduction of
uncertainty: the more numerous the alternatives that are
ruled out, the greater the reduction of uncertainty, and thus
the greater the information. It is usually measured using the
entropy function, which is the logarithm of the number of
alternatives (assuming they are equally likely). For exam-
ple, tossing a fair coin and obtaining heads corresponds to
log2(2) 1 bit of information, because there are just two
alternatives; throwing a fair die yields log2(6) 2.59 bits of
information, because there are six.
Let us now compare the photodiode with you. When the
blank screen turns on, the mechanism in the photodiode tells
the detector that the current from the sensor is above rather
than below the threshold, so it reports light. In performing
this discrimination between two alternatives, the detector in
the photodiode generates log2(2) 1 bit of information.
When you see the blank screen turn on, on the other hand,
217CONSCIOUSNESS AS INTEGRATED INFORMATION
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
3/27
the situation is quite different. Though you may think you
are performing the same discrimination between light and
dark as the photodiode, you are in fact discriminating
among a much larger number of alternatives, thereby gen-
erating many more bits of information.
This is easy to see. Just imagine that, instead of turning
light and dark, the screen were to turn red, then green, then
blue, and then display, one after the other, every frame from
every movie that was ever produced. The photodiode, in-
evitably, would go on signaling whether the amount of light
for each frame is above or below its threshold: to a photo-
diode, things can only be one of two ways, so when it
reports light, it really means just this way versus that
way. For you, however, a light screen is different not only
from a dark screen, but from a multitude of other images, so
when you say light, it really means this specific way
versus countless other ways, such as a red screen, a green
screen, a blue screen, this movie frame, that movie frame,
and so on for every movie frame (not to mention for asound, smell, thought, or any combination of the above).
Clearly, each frame looks different to you, implying that
some mechanism in your brain must be able to tell it apart
from all the others. So when you say light, whether you
think about it or not (and you typically wont), you have just
made a discrimination among a very large number of alter-
natives, and thereby generated many bits of information.
This point is so deceivingly simple that it is useful to
elaborate a bit on why, although a photodiode may be as
good as we are in detecting light, it cannot possibly see light
the way we doin fact, it cannot possibly see anything at
all. Hopefully, by realizing what the photodiode lacks, wemay appreciate what allows us to consciously see the
light.
The key is to realize how the many discriminations we
can do, and the photodiode cannot, affect themeaningof the
discrimination at hand, the one between light and dark. For
example, the photodiode has no mechanism to discriminate
colored from achromatic light, even less to tell which par-
ticular color the light might be. As a consequence, all light
is the same to it, as long as it exceeds a certain threshold. So
for the photodiode, light cannot possibly mean achro-
matic as opposed to colored, not to mention of which
particular color. Also, the photodiode has no mechanism to
distinguish between a homogeneous light and a bright
shapeany bright shapeon a darker background. So for
the photodiode, light cannot possibly mean full field as
opposed to a shapeany of countless particular shapes.
Worse, the photodiode does not even know that it is detect-
ing a visual attribute (the visualness of light) as it has no
mechanism to tell visual attributes, such as light or dark,
from non-visual ones, such as hot and cold, light or heavy,
loud or soft, and so on. As far as it knows, the photodiode
might just as well be a thermistorit has no way of know-
ing whether it is sensing light versusdark or hotversuscold.
In short, the only specification a photodiode can make is
whether things are this or that way: any further specification
is impossible because it does not have mechanisms for it.
Therefore, when the photodiode detects light, such light
cannot possibly mean what it means for us; it does not even
mean that it is a visual attribute. By contrast, when we see
light in full consciousness, we are implicitly being much
more specific: we simultaneously specify that things are this
way rather than that way (light as opposed to dark), that
whatever we are discriminating is not colored (in any par-
ticular color), does not have a shape (any particular one), is
visual as opposed to auditory or olfactory, sensory as op-
posed to thought-like, and so on. To us, then, light is much
more meaningful precisely because we have mechanisms
that can discriminate this particular state of affairs we call
light against a large number of alternatives.
According to the IIT, it is all this added meaning, pro-
vided implicitly by how we discriminate pure light from all
these alternatives, that increases the level of consciousness.This central point may be appreciated either by subtrac-
tion or by addition. By subtraction, one may realize that
our being conscious of light would degrade more and
morewould lose its non-coloredness, its non-shapedness,
would even lose its visualnessas its meaning is progres-
sively stripped down to just one of two ways, as with the
photodiode. By addition, one may realize that we can only
see light as we see it, as progressively more and more
meaning is added by specifying how it differs from count-
less alternatives. Either way, the theory says that the more
specifically ones mechanisms discriminate between what
pure light is and what it is not (the more they specify whatlight means), the more one is conscious of it.
Integration: the camera thought experiment
Informationthe ability to discriminate among a large
number of alternativesmay thus be essential for con-
sciousness. However, information always implies a point of
view, and we need to be careful about what that point of
view might be. To see why, consider another thought ex-
periment, this time involving a digital camera, say one
whose sensor chip is a collection of a million binary pho-
todiodes, each sporting a sensor and a detector. Clearly,
taken as a whole, the cameras detectors could distinguish
among 21,000,000 alternative states, an immense number,
corresponding to 1 million bits of information. Indeed, the
camera would easily respond differently to every frame
from every movie that was ever produced. Yet few would
argue that the camera is conscious. What is the key differ-
ence between you and the camera?
According to the IIT, the difference has to do with
integrated information. From the point of view of an exter-
nal observer, the camera may be considered as a single
system with a repertoire of 21,000,000 states. In reality, how-
218 G. TONONI
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
4/27
ever, the chip is not an integrated entity: since its 1 million
photodiodes have no way to interact, each photodiode per-
forms its own local discrimination between a low and a high
current completely independent of what every other photo-
diode might be doing. In reality, the chip is just a collection
of 1 million independent photodiodes, each with a repertoire
of two states. In other words, there is no intrinsic point of
view associated with the camera chip as a whole. This is
easy to see: if the sensor chip were cut into 1 million pieces
each holding its individual photodiode, the performance of
the camera would not change at all.
By contrast, you discriminate among a vast repertoire of
states as an integrated system, one that cannot be broken
down into independent components each with its own sep-
arate repertoire. Phenomenologically, every experience is
an integrated whole, one that means what it means by virtue
of being one, and that is experienced from a single point of
view. For example, the experience of a red square cannot be
decomposed into the separate experience of red and theseparate experience of a square. Similarly, experiencing the
full visual field cannot be decomposed into experiencing
separately the left half and the right half: such a possibility
does not even make sense to us, since experience is always
whole. Indeed, the only way to split an experience into
independent experiences seems to be to split the brain in
two, as in patients who underwent the section of the corpus
callosum to treat severe epilepsy (Gazzaniga, 2005). Such
patients do indeed experience the left half of the visual field
independently of the right side, but then the surgery has
created two separate consciousnesses instead of one. Mech-
anistically then, underlying the unity of experience must becausal interactions among certain elements within the brain.
This means that these elements work together as an inte-
grated system, which is why their performance, unlike that
of the camera, breaks down if they are disconnected.
A Mathematical Analysis: Quantifying Integrated
Information
This phenomenological analysis suggests that, to gener-
ate consciousness, a physical system must be able to dis-
criminate among a large repertoire of states (information)
and it must be unified; that is, it should be doing so as a
single system, one that is not decomposable into a collectionof causally independent parts (integration). But how can one
measure integrated information? As I explain below, the
central idea is to quantify the information generated by a
system, above and beyond the information generated inde-
pendently by its parts (Tononi, 2001, 2004; Balduzzi and
Tononi, 2008).1
Information
First, we must evaluate how much information is gener-
ated by the system. Consider the system of two binary units
in Figure 1, which can be thought of as an idealized version
of a photodiode composed of a sensor S and a detector D.
The system is characterized by a state it is in, which in this
case is 11 (first digit for the sensor, second digit for the
detector), and by a mechanism. This is mediated by a
connection (arrow) between the sensor and the detector that
implements a causal interaction: in this case, the elementary
mechanism of the system is that the detector checks the state
of the sensor and turns on if the sensor is on, and off
otherwise (more generally, the specific causal interaction
can be described by an input-output table).
Potentially, a system of two binary elements could be in
any of four possible states (00,01,10,11) with equal proba-
1 2
SENSOR DETECTOR
P
1/4
0 0 1 10 1 0 1
P
1/2
0 0 1 10 1 0 1
A.
B.
1
2
ei(X(mech,x1)) = H [p(X
0(mech, x
1)) ||p(X
0(maxH))] = 1 bit
p(X0(maxH))
p(X0(mech, x
1))
Figure 1. Effective information. (A) A photodiode consisting of a
sensor and detector unit. The photodiodes mechanism is such that the detector
unit turns on if the sensors current is above a threshold. Here both units are on
(binary 1, indicated in gray). (B) For the entire system (sensor unit, detector
unit) there are four possible states: (00,01,10,11). The potential distribution
p(X0(maxH)) (1/4,1/4,1/4,1/4) is the maximum entropy distribution on the
four states. Given the photodiodes mechanism and the fact that the detector is
on, the sensor must have been on. Thus, the photodiodes mechanism and its
current state specifies the following distribution: two of the four possible states
(00,01) are ruled out; the other two states (10,11) are equally likely since they
areindistinguishableto themechanism (the priorstate of the detector makes no
difference to the current state of the sensor). The actual distribution is therefore
p(X0(mech, x1)) (0,0,1/2,1/2). Relative entropy (Kullback-Leibler diver-
gence) between two probability distributions pandqis H[p|q] pilog2pi/qi,
so the effective information ei(X(mech, x1)) associated with output x1 11 is
1 bit (effective information is the entropy of the actual relative to the potential
distributions).
219CONSCIOUSNESS AS INTEGRATED INFORMATION
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
5/27
bility: p (1/4,1/4,1/4,1/4). Formally, this potential (a
priori) repertoire is represented by the maximum entropy or
uniform distribution of possible system states at time t0,
which expresses complete uncertainty (p(X0(maxH))). Con-
sidering the potential repertoire as the set of all possible
input states, the particular mechanism X(mech) of this sys-
tem can be thought of as specifying a forwardrepertoire
the probability distribution of output states produced by the
system when perturbed with all possible input states. But the
system is actually in a particular output state (in this case, at
time t1, x1 11). In actuality, a system with this mech-
anism being in state 11 specifies that the previous system
state x0must have been either 11 or 10, rather than 00 or 01,
corresponding to p (0,0,1/2,1/2) (in this system, there is
no mechanism to specify the detector state, which remains
uncertain). Formally, then, the mechanism and the state 11
specify anactual (a posteriori) distribution or repertoire of
system states p(X0(mech,x1)) at time t0 that could have
caused (led to) x1 at time t1, while ruling out (givingprobability zero to) states that could not. In this way, the
systems mechanism and state constitute information (about
the systems previous state), in the classic sense of reduction
of uncertainty or ignorance. More precisely, the systems
mechanism and state generate 1 bit of information by dis-
tinguishing between things being one way (11 or 10, which
remain indistinguishable to it) rather than another way (00
or 01, which also remain indistinguishable to it).
In general, the information generated when a system
characterized by a certain mechanism in a particular state
can be measured by the relative entropy H between the
actual and the potential repertoires (relative to is indicatedby ), captured by the effective information (ei):
eiXmech,x1 HpX0mech,x1pX0maxH
Relative entropy, also known as Kullback-Leibler diver-
gence, is a difference between probability distributions
(Cover and Thomas, 2006): if the distributions are identical,
relative entropy is zero; the more different they are, the
higher the relative entropy.2 Figuratively, the systems
mechanism and state generate information by sharpening
the uniform distribution into a less uniform onethis is
how much uncertainty is reduced. Clearly, the amount of
effective information generated by a system is high if it has
a large potential repertoire and a small actual repertoire,
since a large number of initial states are ruled out. By
contrast, the information generated is little if the systems
repertoire is small, or if many states could lead to the current
outcome, since few states are ruled out. For instance, if
noise dominates (any state could have led to the current
one), no alternatives are ruled out, and no information is
generated.
Since effective information is implicitly specified once a
mechanism and state are specified, it can be considered to be
an intrinsic property of a system. To calculate it explic-
itly, from an extrinsic perspective, one can perturb the
system in all possible ways (i.e., try out all possible input
states, corresponding to the maximum entropy distribution
or potential repertoire) to obtain the forward repertoire of
output states given the systems mechanism. Finally one can
calculate, using Bayes rule, the actual repertoire given the
systems state (Balduzzi and Tononi, 2008).3
Integration
Second, we must find out how much of the information
generated by a system is integrated information; that is, how
much information is generated by a single entity, as opposed
to a collection of independent parts. The idea here is to
consider the parts of the system independently, ask how
much information they generate by themselves, and compare it
with the information generated by the system as a whole.
This can be done by resorting again to relative entropy tomeasure the difference between the probability distribution
generated by the system as a whole (p(X0(mech,x1)), the
actual repertoire of the system x) with the probability dis-
tribution generated by the parts considered independently
(p(kM0(mech,1)), the product of the actual repertoire of
the parts kM). Integrated information is indicated with the
symbol (the vertical bar I stands for information, the
circle O for integration):
Xmech,x1
HpX0mech,x1pkM0mech,1for
kM0 MIP
That is, the actual repertoire for each part is specified by
causal interactions internal to each part, considered as a
system in its own right, while external inputs are treated as
a source of extrinsic noise. The comparison is made with the
particular decomposition of the system into parts that leaves
the least information unaccounted for. This minimum infor-
mation partition (MIP) decomposes the system into its
minimal parts.
To see how this works, consider two of the million
photodiodes in the digital camera (Fig. 2, left). By turning
on or off depending on its input, each photodiode generates
1 bit of information, just as we saw before. Considered
independently, then, two photodiodes generate 2 bits of
information, and 1 million photodiodes generate 1 million
bits of information. However, as shown in the figure, the
product of the actual distributions generated independently
by the parts is identical to the actual distribution for the
system. Therefore, the relative entropy between the two
distributions is zero: the system generates no integrated
information ( (X(mech,x1)) 0) above and beyond what
is generated by its parts.
Clearly, for integrated information to be high, a system
must be connected in such a way that information is gen-
220 G. TONONI
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
6/27
erated by causal interactions among rather than within its
parts. Thus, a system can generate integrated information
only to the extent that it cannot be decomposed into infor-
mationally independent parts. A simple example of such a
system is shown in Figure 2 (right). In this case, the inter-
action between the minimal parts of the system generates
information above and beyond what is accounted for by the
parts by themselves ( (X(mech,x1)) 0).
In short, integrated information captures the information
generated by causal interactions in the whole, over and
above the information generated by the parts.4
Complexes
Finally, by measuring values for all subsets of elements
within a system, we can determine which subsets form
complexes. Specifically, a complex X is a set of elements
that generate integrated information ( 0) that is not fully
contained in some larger set of higher
(Fig. 3). A com-plex, then, can be properly considered to form a single
entity having its own, intrinsic point of view (as opposed
to being treated as a single entity from an outside, extrinsic
point of view). Since integrated information is generated
withina complex and not outside its boundaries, experience
is necessarily private and related to a single point of view or
perspective (Tononi and Edelman, 1998; Tononi, 2004). A
given physical system, such as a brain, is likely to contain
more than one complex, many small ones with low
values, and perhaps a few large ones (Tononi and Edelman,
1998; Tononi, 2004). In fact, at any given time there may be
a singlemain complexof comparatively much higher that
underlies the dominant experience (a main complex is suchthat its subsets have strictly lower ). As shown in Figure
3, a main complex can be embedded into larger complexes
of lower . Thus, a complex can be casually connected,
throughports-inand ports-out, to elements that are not part
of it. According to the IIT, such elements can indirectly
influence the state of the main complex without contributing
directly to the conscious experience it generates (Tononi
and Sporns, 2003).
A Neurobiological Reality Check: Accounting for
Empirical Observations
Can this approach account, at least in principle, for some
of the basic facts about consciousness that have emerged
from decades of clinical and neurobiological observations?
Measuring and finding complexes is not easy for realistic
systems, but it can be done for simple networks that bear
some structural resemblance to different parts of the brain
(Tononi, 2004; Balduzzi and Tononi, 2008).
For example, by using computer simulations, it is possi-
ble to show that high requires networks that conjoin
functional specialization (due to its specialized connectiv-
ity; each element has a unique functional role within the
network) with functional integration (there are many path-
ways for interactions among the elements, Fig. 4A.). In very
rough terms, this kind of architecture is characteristic of the
mammalian corticothalamic system: different parts of the
cerebral cortex are specialized for different functions, yet a
vast network of connections allows these parts to interact
profusely. And indeed, as much neurological evidence in-
dicates (Posner and Plum, 2007), the corticothalamic system
is precisely the part of the brain that cannot be severely
impaired without loss of consciousness.
Conversely, is low for systems that are made up of
small, quasi-independent modules (Fig. 4B; Tononi, 2004;
Balduzzi and Tononi, 2008). This may be why the cerebel-
lum, despite its large number of neurons, does not contrib-
ute much to consciousness: its synaptic organization is such
that individual patches of cerebellar cortex tend to be acti-
vated independently of one another, with little interaction
between distant patches (Bower, 2002).
Computer simulations also show that units along multi-ple, segregated incoming or outgoing pathways are not
incorporated within the repertoire of the main complex (Fig.
4C; Tononi, 2004; Balduzzi and Tononi, 2008). This may
be why neural activity in afferent pathways (perhaps as far
as V1), though crucial for triggering this or that conscious
experience, does not contribute directly to conscious expe-
rience; nor does activity in efferent pathways (perhaps start-
ing with primary motor cortex), though it is crucial for
reporting each different experience.
The addition of many parallel cycles also generally does
not change the composition of the main complex, although
values can be altered (Fig. 4D). Instead, cortical andsubcortical cycles or loops implement specialized subrou-
tines that are capable of influencing the states of the main
corticothalamic complex without joining it. Such informa-
tionally insulated cortico-subcortical loops could constitute
the neural substrates for many unconscious processes that
can affect and be affected by conscious experience (Baars,
1988; Tononi, 2004), such as those that enable object rec-
ognition, language parsing, or translating our vague inten-
tions into the right words.
At this stage, it is hard to say precisely which cortical
circuits may work as a large complex of high , and which
instead may remain informationally insulated. Does the
dense mesial connectivity revealed by diffusion spectral
imaging (Hagmann et al., 2008) constitute the backbone
of a corticothalamic main complex? Do parallel loops
through basal ganglia implement informationally insulated
subroutines? Are primary sensory cortices organized like
massive afferent pathways to a main complex higher up in
the cortical hierarchy (Koch, 2004)? Is much of prefrontal
cortex organized like a massive efferent pathway? Do cer-
tain cortical areas, such as those belonging to the dorsal
visual stream, remain partly segregated from the main com-
plex? Unfortunately, answering these questions and prop-
221CONSCIOUSNESS AS INTEGRATED INFORMATION
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
7/27
INTEGRATED INFORMATION GENERATED BY THE SYSTEM ABOVE AND BEYOND THE PARTS
INFORMATION GENERATED BY THE SYSTEM
INFORMATION GENERATED BY THE PARTS
A
1
2
3
4
P
1/4
1/16
P
3/8
B
1/41/4
P
2/3
1/4
P
1/2
B
1
2
3
4
P
1/2
1/4
A P 1
1/16
C
1
2
3
4
P
1/4
1/4
C P 1
1/4
1
2
3
4
1
2
3
4
31
2 4
MIP
ei(X(mech,x1)) = 2 bits ei(X(mech,x
1)) = 4 bits
actual: p(X0(mech,x1))
potential: p(X0(maxH))
actual: p(X0(mech,x
1))
potential: p(X0(maxH))
ei(aM(mech,1))=1 bit
aM bM
aM bM
p(aM0(mech,
1))
aM bM
MIP
aM bM
MIP MIPp(kM
0(mech,
1))
K=1,2 K=1,2
(X(mech,x1))=H[p(X
0(mech,x
1))||p(kM
0(mech,
1))]=0 bits
K=1,2 K=1,2
p(bM0(mech,
1))
p(aM0(maxH)) p(bM0(maxH))
ei(bM(mech,1))=1 bit
p(aM0(mech,
1)) p(bM0(mech,1))
p(aM0(maxH)) p(bM0(maxH))
ei(aM(mech,1))=1.1 bits ei(bM(mech,
1))=1 bit
p(X0(mech,x
1)) p(X
0(mech,x
1))
p(kM0(mech,
1))
(X(mech,x1))=H[p(X
0(mech,x
1))||p(kM
0(mech,
1))]=2 bits
Figure 2. Integrated information. Left-hand side: two photodiodes in a digital camera. (A) Information
generated by the system as a whole. The system as a whole generates 2 bits of effective information by
specifying that n1 and n3must have been on. (B) Information generated by the parts. The minimum information
partition (MIP) is the decomposition of a system into (minimal) parts, that is, the decomposition that leaves the
least information unaccounted for. Here the parts are two photodiodes. (C) The information generated by the
system as a whole is completely accounted for by the information generated by its parts. In this case, the actual
repertoire of the whole is identical to the combined actual repertoires of the parts (the product of their
222 G. TONONI
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
8/27
erly testing the predictions of the theory requires a much
better understanding of cortical neuroanatomy than is cur-
rently available.
Other simulations show that the effects of cortical dis-
connections are readily captured in terms of integrated
information (Tononi, 2004): a callosal cut produces, out
of a large complex corresponding to the connected cortico-
thalamic system, two separate complexes, in line with many
studies of split-brain patients (Gazzaniga, 2005). However,
because there is great redundancy between the two hemi-
spheres, their value is not greatly reduced compared to
when they form a single complex. Functional disconnec-
tions may also lead to a restriction of the neural substrate of
consciousness, as is seen in neurological neglect phenom-
ena, in psychiatric conversion and dissociative disorders,
and possibly during dreaming and hypnosis. It is also likely
that certain attentional phenomena may correspond to
changes in the composition of the main complex underlying
consciousness (Koch and Tsuchiya, 2007). The attentionalblink,5 where a fixed sensory input may at times make it to
consciousness and at times not, may also be due to changes
in functional connectivity: access to the main corticotha-
lamic complex may be enabled or not based on dynamics
intrinsic to the complex (Dehaene et al., 2003). Similarly,
binocular rivalry6 may be related, at least in part, to dy-
namic changes in the composition of the main corticotha-
lamic complex caused by transient changes in functional
connectivity. Computer simulations confirm that functional
disconnection can reduce the size of a complex and reduce
its capacity to integrate information (Tononi, 2004). While
it is not easy to determine, at present, whether a particulargroup of neurons is excluded from the main complex
because of hard-wired anatomical constraints or is tran-
siently disconnected due to functional changes, the set of
elements underlying consciousness is not static, but form
a dynamic complex o r dynamic core (Tononi and
Edelman, 1998).
Computer simulations also indicate that the capacity to
integrate information is reduced if neural activity is ex-
tremely high and near-synchronous, due to a dramatic de-
crease in the repertoire of discriminable states (Fig. 4E;
Balduzzi and Tononi, 2008). This reduction in degrees of
freedom could be the reason that consciousness is reduced
or eliminated in absence seizure (petit mal) and other con-
ditions during which neural activity is both high and syn-
chronous (Blumenfeld and Taylor, 2003).
The most common example of a marked change in the
level of experience is the fading of consciousness that
occurs during certain periods of sleep. Subjects awakened in
deep NREM (nonrapid eye movement) sleep, especially
early in the night, often report that they were not aware of
themselves or of anything else, though cortical and thalamic
neurons remain active. Awakened at other times, mainly
during REM sleep or during lighter periods of NREM sleep
later in the night, they report dreams characterized by vivid
images (Hobson et al., 2000). From the perspective of
integrated information, a reduction of consciousness during
early sleep would be consistent with the bistability of cor-
tical circuits during deep NREM sleep. Due to changes in
intrinsic and synaptic conductances triggered by neuro-
modulatory changes (e.g., low acetylcholine), cortical neu-
rons cannot sustain firing for more than a few hundred
milliseconds and invariably enter a hyperpolarized down-
state. Shortly afterward, they inevitably return to a depolar-
ized up-state (Steriade et al., 2001). Indeed, computer sim-ulations show that values of are low in systems with such
bistable dynamics (Fig. 4F, Balduzzi and Tononi, 2008).
Consistent with these observations, studies using TMS, a
technique for stimulating the brain non-invasively, in con-
junction with high-density EEG, show that early NREM
sleep is associated either with a breakdown of the effective
connectivity among cortical areas, and thereby with a loss of
integration (Massiminiet al., 2005, 2007), or with a stereo-
typical global response suggestive of a loss of repertoire and
thus of information (Massimini et al., 2007). Similar
changes are seen in animal studies of anesthesia (Alkire et
al., 2008).Finally, consciousness not only requires a neural sub-
strate with appropriate anatomical structure and appropriate
physiological parameters, it also needs time (Bachmann,
2000). The theory predicts that the time requirement for the
generation of conscious experience in the brain emerges
directly from the time requirements for the build-up of an
integrated repertoire among the elements of the corticotha-
lamic main complex so that discriminations can be highly
informative (Tononi, 2004; Balduzzi and Tononi, unpubl.).
To give an obvious example, if one were to perturb half of
the elements of the main complex for less than a millisec-
ond, no perturbations would produce any effect on the other
half within this time window, and would be zero. After,
say, 100 ms, however, there is enough time for differential
effects to be manifested, and should grow.
respective probability distributions), so that relative entropy is zero. The system generates no information above and beyond the parts, so it cannot be
considered a single entity.Right-hand side: an integrated system. Elements in the system are on if they receive two or more spikes. The system is in state
x1 1000. (A) The mechanism specifies a unique prior state that can cause state x1, so the system generates 4 bits of effective information. All other initial
states are ruled out, since they cause different outputs. (B) Effective information generated by the two minimal parts, considered as systems in their own
right. External inputs are treated as extrinsic noise. (C) Integrated information is information generated by the whole (black arrows) over and above the
parts (gray arrows). In this case, the actual repertoire of the whole is different from the combined actual repertoires of the parts, and the relative entropy
is 2 bits. The system generates information above and beyond the parts, so it can be considered a single entity (a complex).
223CONSCIOUSNESS AS INTEGRATED INFORMATION
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
9/27
The Quality of Consciousness: Characterizing
Informational Relationships
If the amount of integrated information generated by
different brain structures (or by the same structure function-
ing in different ways) can in principle account for changesin the level of consciousness, what is responsible for the
quality of each particular experience? What determines that
colors look the way they do and are different from the way
music sounds? Once again, empirical evidence indicates
that different qualities of consciousness must be contributed
by different cortical areas. Thus, damage to certain parts of
the cerebral cortex forever eliminates our ability to experi-
ence color (whether perceived, imagined, remembered, or
dreamt), whereas damage to other parts selectively elimi-
nates our ability to experience visual shapes. There is ob-
viously something about different parts of the cortex that
can account for their different contribution to the quality of
experience. What is this something?
The IIT claims that, just as thequantityof consciousness
generated by a complex of elements is determined by the
amount of integrated information it generates above and
beyond its parts, the quality of consciousness is determined
by the set of all the informational relationships its mecha-
nisms generate. That is,how integrated information is gen-
erated within a complex determines not only the amount of
consciousness it has, but also what kind of consciousness.
Consider again the photodiode thought experiment. As I
discussed before, when the photodiode reacts to light, it can
only tell that things are one way rather than another way. On
the other hand, when we see light, we discriminate against
many more states of affairs, and thus generate much more
information. In fact, I argued that light means what it
means and becomes conscious lightby virtue ofbeing not
just the opposite of dark, but also different from any color,
any shape, any combination of colors and shapes, any frame
of every possible movie, any sound, smell, thought, and so on.
What needs to be emphasized at this point is that dis-
criminating light against all these alternatives implies not
just picking one thing out of everything else (an undif-
ferentiated bunch), but distinguishing at once, in a specific
way, between each and every alternative. Consider a very
simple example: a binary counter capable of discriminating
among the four numbers: 00, 01, 10, 11. When the counter
says binary 3, it is not just discriminating 11 from every-
thing else as an undifferentiated bunch, otherwise it would
not be a counter, but a 11 detector. To be a counter, the
system must be able to tell 11 apart from 00 as well as from10 as well as from 01 in different, specific ways. It does so,
of course, by making choices through its mechanisms; for
example: is this the first or the second digit? Is it a 0 or a 1?
Each mechanism adds its specific contribution to the dis-
crimination they perform together. Similarly, when we see
light, mechanisms in our brain are not just specifying light
with respect to a bunch of undifferentiated alternatives.
Rather, these mechanisms are specifying that light is what it
is by virtue of being different, in this and that specific way,
from every other alternativefrom dark to any color, to any
shape, movie frame, sound or smell, and so on.
In short, generating a large amount of integrated infor-mation entails having a highly structured set of mechanisms
that allow us to make many nested discriminations (choices)
as a single entity. According to the IIT, these mechanisms
working together generate integrated information by speci-
fying a set of informational relationships that completely
and univocally determine the quality of experience.
Experience as a shape in qualia space
To see how this intuition can be given a mathematical
formulation, let us consider again a complex of n binary
elements X(mech,x1) having a particular mechanism and
being in a particular state. The mechanism of the system is
implemented by a set of connections Xconn among its ele-
ments. Let us now suppose that each possible state of the
system constitutes an axis or dimension of a qualia space
(Q) having 2n dimensions. Each axis is labeled with the
probability p for that state, going from 0 to 1, so that a
repertoire (i.e., a probability distribution on the possible
states of the complex) corresponds to a point in Q (Fig. 5).
Let us now examine how the connections among the
elements of the complex specify probability distributions;
that is, how a set of mechanisms specifies a set of informa-
(x1)=
2
(a1)=
3
(b1)=
1
(s1)=
2
(x)= 11
()= 3
(s) = 1
() = 2
Figure 3. Complexes. In this system, the mechanism is that elements
fire in response to an odd number of spikes on their afferent connections
(links without arrows are bidirectional connections). Analyzing the systemin terms of integrated information shows that the system constitutes a
complex (x, light gray) that contains three smaller complexes (s,a,b, in
different shades of gray). Observe that (i) complexes can overlap; (ii) a
complex can interact causally with elements not part of it; (iii) groups of
elements with identical architectures (a and b) generate different amounts
of integrated information, depending on their ports-in and ports-out.
224 G. TONONI
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
10/27
0 2 4 6 80
1
2
3
4
Max
= 3.7
= .17
= 0
Elements firing
COMATOSE, BALANCED & EPILEPTIC SYSTEMS SLEEPING SYSTEM
= 4
= 3.6
= 1.9
= 3.6
= 1
CORTICOTHALAMIC SYSTEM
AFFERENT PATHWAYS CORTICAL-SUBCORTICAL LOOPS
A
f = 1.3
= .4
CEREBELLAR SYSTEM
f = 1.8
= 1.8
B
C D
time (ticks)
100
%a
ctivity
0
50
2
0
1
% active
0 20 40 60
INTEGRATED INFORMATION & NEUROANATOMY
E F
INTEGRATED INFORMATION & NEUROPHYSIOLOGY
Figure 4. Relating integrated information to neuroanatomy and neurophysiology. Elements fire in
response to two or more spikes (except elements targeted by a single connection, which copy their input); links
without arrows are bidirectional. (A) Computing in simple models of neuroanatomy suggests that a
functionally integrated and functionally specialized networklike the corticothalamic systemis well suited to
generating high values of . (B, C, D) Architectures modeled on the cerebellum, afferent pathways, and
cortical-subcortical loops give rise to complexes containing more elements, but with reduced compared to the
main corticothalamic complex. (E) peaks in balanced states; if too many or too few elements are active,
collapses. (F) In a bistable (sleeping) system (same as in (E)), collapses when the number of firing elements
(dotted line) is too high (high % activity), remains low during the DOWN state (zero % activity), and only
recovers at the onset of the next UP state.
225CONSCIOUSNESS AS INTEGRATED INFORMATION
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
11/27
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
12/27
formational relationship can be represented as an arrow in Q
(q-arrow) that goes from the point corresponding to the
maximum entropy distribution (p 1/2n) to the point cor-
responding to the actual repertoire specified by that connec-
tion. The length (divergence) of the q-arrow expresseshow
much the connection specifies the distribution (the effective
information it generates, i.e., the relative entropy between
the two distributions); the direction in Q expresses the
particular way in which the connection specifies the distri-
bution, i.e., a change in position in Q. Similarly, if one
considers all other connections taken in isolation, each will
specify another q-arrow of a certain length, pointing in a
different direction.
Next, consider all possible combinations of connections
(Fig. 5B). For instance, consider adding the contribution of
the second connection to that of the first. Together, the first
and second connections specify another actual repertoire
another point in Q-spaceand thereby generate more in-
formation than either connection alone as they shape theuniform distribution into a more specific distribution. To the
tip of the q-arrow specified by the first connection, one can
now add a q-arrow bent in the direction contributed by the
second connection, forming an edge of two q-arrows in
Q-space (the same final point is reached by adding the
q-arrow due to the first connection on top of the q-arrow
specified by the second one). Each combination of connec-
tion therefore specifies a q-edge made of concatenated q-
arrows (component q-arrows). In general, the more connec-
tions one considers together, the more the actual repertoire
will take shape and differ from the uniform (potential)
distribution.Finally, consider the joint contribution of all connections
of the complex (Fig. 5B). As was discussed above, all
connections together specify the actual repertoire of the
whole. This is the point where all q-edges converge. To-
gether, these q-edges in Q delimit a quale, that is, a shape
in Q, a kind of 2n-dimensional solid (technically, in more
than three dimensions, the body of a polytope). The
bottom of the quale is the maximum entropy distribution, its
edges are q-edges made of concatenated q-arrows, and its
top is the actual repertoire of the complex as a whole. The
shape of this solid (polytope) is specified by all informa-
tional relationships that are generated within the complex by
the interactions among its elements (the effective informa-
tion matrix; Tononi, 2004).7 Note that the same complex of
elements, endowed with the same mechanism, will typically
generate a different quale or shape in Q depending on the
particular state it is in.
It is worth considering briefly a few relevant properties of
informational relationships or q-arrows. First, informational
relationships are context-dependent (Fig. 6), in the follow-
ing sense. Acontextcan be any point in Q corresponding to
the actual repertoire generated by a particular subset of
connections. It can be shown that the q-arrow generated by
considering the effects of an additional connection (how it
further sharpens the actual repertoire) can change in both
magnitude and direction depending on the context in which
it is considered. In Figure 6, when considered in isolation
(null context), the connection r between elements 4 and 3
generates a short q-arrow (0.18 bits) pointing in a certain
direction. When considered in the full context provided by
all other connections (not-r or r), the same connection r
generates a longer q-arrow (1 bit) pointing in a different
direction.
Another property is how removing or adding a set of
connections folds or unfolds a quale. The portion of the
quale that is generated by a set of connections r (acting in all
contexts) is called aq-fold. If we remove connection r from
the system, all the q-arrows generated by that connection, in
all possible contexts, vanish, so the shape of the quale
folds along the q-fold specified by that connection. Con-
versely, when the connection is added to a system, the shape
of the quale unfolds.Another important property of q-arrows is entanglement
(, Balduzzi and Tononi, unpubl.). A q-arrow is entangled
( 0) if the underlying connections considered together
generate information above and beyond the information
they generate separately (note the analogy with ). Thus,
entanglement characterizes informational relationships (q-
arrows) that are more than the sum of their component
relationships (component q-arrows, Fig. 6B), just like
characterizes systems that are more than the sum of their
parts. Geometrically, entanglement warps the shape of the
quale away from a simple hypercube (where q-arrows are
orthogonal to each other). Entanglement has several rele-vant consequences (Balduzzi and Tononi, unpubl.). For
example, an entangled q-arrow can be said to specify a
concept, in that it groups together certain states of affairs in
a way that cannot be decomposed into the mere sum of
simpler groupings (see also Feldman, 2003). Moreover, just
as can be used to identify complexes, entanglement can
be used to identify modes. By analogy with complexes,
modesare sets of q-arrows that are more densely entangled
than surrounding q-arrows: they can be considered as clus-
ters of informational relationships constituting distinctive
sub-shapes in Q (see Fig. 8). By analogy with a main
complex, an elementary mode is such that its component
q-arrows have strictly lower . As will be briefly discussed
below, modes play an important role in understanding the
structure of experience.
Some properties of qualia space
What is the relevance of these constructs to understand-
ing the quality of consciousness? It is not easy to become
familiar with a complicated multidimensional space nearly
impossible to draw, so it may be useful to resort to some
metaphors. I have argued that the set of informational rela-
227CONSCIOUSNESS AS INTEGRATED INFORMATION
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
13/27
tionships in Q generated by the mechanisms of a complex in
a given state (q-arrows between repertoires) specify a shape
in Q (a quale). Perhaps the most important notion emerging
from this approach is that an experience is a shape in Q.
According to the IIT, this shape completely and univo-
cally8 specifies the quality of experience.
It follows that different experiences are, literally, differ-
ent shapes in Q. For example, when the same system is in a
different state (firing pattern), it will typically generate a
different shape or quale (even for the same value of ).
Importantly, if an element turns on, it generates information
and meaning not by signifying something (say red),
which in isolation it cannot, but by changing the shape of
the quale. Moreover, experiences are similar if their shape is
similar, and different to the extent that their shapes are
different. This means that phenomenological similarities
and differences can in principle be quantified as similarities
and differences between shapes. The set of all shapes gen-
erated by the same system in different states provides a
geometrical depiction of all its possible experiences.9
Note that a quale can only be specified by a mechanism
and a particular stateit does not make sense to ask about
the quale generated by a mechanism in isolation, or by a
state (firing pattern) in isolation. A consequence is that two
different systems in the same state can generate two differ-
ent experiences (i.e., two different shapes). As an extreme
example, a system that was to copy one by one the state of
the neurons in a human brain, but had no internal connec-
tions of its own, would generate no consciousness and no
quale (Tononi, 2004; Balduzzi and Tononi, 2008).
By the same token, it is possible that two different sys-
tems generate the same experience (i.e., the same shape).
.18 bits
1 bit
entanglement = .42 bits
r
r
r
r
r
r
1
2
3
4
1
2
3
4
1
2
3
4
1
2
3
4
B
A
NULL CONTEXTFULL CONTEXT
Figure 6. Context and entanglement. (A) Context. The same connection (black arrow between elements
3 and 4) considered in two contexts. At the bottom of the quale (null context, corresponding to the maximum
entropy distribution when no other connections are engaged), the connection r generates a q-arrow (called
down-set of r, or2r) corresponding to 0.18 bits of information pointing up-left in Q. Near the top of the quale(full context, corresponding to the actual distribution specified by all other connections except for r, indicated
as r), r generates a q-arrow (called up-set of non-red, or 1 r) corresponding to 1 bit of information pointingup-right in Q. (B) Entanglement. Left: the q-arrow generated by the connection r and the q-arrow generated
by the complementary connections r at the bottom of the quale (null context). Right: The product of the twoq-arrows (corresponding to independence between the informational relationships specified by the two sets of
connections) would be a point corresponding to the vertex of the dotted parallelogram opposite to the bottom.
However, r and r jointly specify the actual distribution corresponding to the top of the quale (black
triangle). The distance between the probability distribution in Q specified jointly by two sets of connections and
their product distribution (zigzag arrow) is the entanglement between the two corresponding q-arrows (how
much the composite q-arrow specifies above and beyond its component q-arrows).
228 G. TONONI
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
14/27
For example, consider again the photodiode, whose mech-
anism determines that if the current in the sensor exceeds a
threshold, the detector turns on. This simple causal interac-
tion is all there is, and when the photodiode turns on it
merely specifies an actual repertoire where states
(00,01,10,11) have, respectively, probability (0,0,1/2,1/2).
This corresponds in Q to a single q-arrow, one bit long,
going from the potential, maximum entropy repertoire (1/
4,1/4,1/4,1/4) to (0,0,1/2,1/2). Now imagine the light sensor
is substituted by a temperature sensor with the same thresh-
old and dynamic rangewe have a thermistor rather than a
photodiode. Although the physical device has changed,
according to the IIT the experience, minimal as it is, has to
be the same, since the informational relationship that is
generated by the two devices is identical. Similarly, an
AND gate when silent and an OR gate when firing also
generate the same shape in Q, and therefore must generate
the same minimal experience (it can be shown that the two
shapes are isomorphic, that is, have the same symmetries;Balduzzi and Tononi, unpubl.). In other words, different
physical systems (possibly in different states) generate the
same experience if the shape of the informational relation-
ships they specify is the same. On the other hand, more
complex networks of causal interactions are likely to create
highly idiosyncratic shapes, so systems of high are un-
likely to generate exactly identical experiences.
If experience is integrated information, it follows that
only the informational relationships within a complex (those
that give the quale its shape) contribute to experience.
Conversely, the informational relationships that exist out-
side the main complexfor example, those involving sen-sory afferents or cortico-subcortical loops implementing
informationally insulated subroutinesdo not make it into
the quale, and therefore do not contribute either to the
quantity or to the quality of consciousness.
Note also that informational relationships, and thus the
shape of the quale, are specified both by the elements that
are firing and by those that are not. This is natural consid-
ering that an element that does not fire will typically rule out
some previous states of affairs (those that would have made
it fire), and thereby it will contribute to specifying the actual
repertoire. Indeed, many silent elements can rule out, in
combination, a vast number of previous states and thus be
highly informative. From a neurophysiological point of
view, such a corollary may lead to counterintuitive predic-
tions. For example, take elements (neurons) within the main
complex that happen to be silent when one is having a
particular experience. If one were to temporarily disable
these neurons (e.g., make them incapable of firing), the
prediction is that, though the system state (firing pattern)
would remain the same, the quantity and quality of experience
would change (Tononi, 2004; Balduzzi and Tononi, 2008).
It is important to see what corresponds to in this
representation (Fig. 7A). The minimum information parti-
tion (MIP) is just another point in Q: the one specified by
the connections within the minimal parts only, leaving out
the contribution of the connections among the parts. This
point is the actual repertoire corresponding to the product of
the actual repertoires of the parts taken independently.
corresponds then to an arrow linking this point to the top of
the solid. In this view, the q-edges leading to the minimum
information bipartition provide the natural base upon
which the solid reststhe informational relationships gen-
erated within the parts upon which are built the informa-
tional relationships amongthe parts. The -arrow can then
be thought of as the height of the solidor rather, to
employ a metaphor, as the highest pole holding up a tent.
For example, if is zero (say a system decomposes into
two independent complexes as in Fig. 7B), the tent corre-
sponding to the system is flatit has no shapesince the
actual repertoire of the system collapses onto its base (MIP).
This is precisely what it means when 0. Conversely,
the higher the value of a complex (the higher the tent orsolid), the more breathing room there is for the various
informational relationships within the complex (the edges of
the solid or the seams of the tent) to express themselves.
In summary, and not very rigorously, the generation of an
experience can be thought of as the erection of a tent with
a very complex structure: the edges are the tension lines
generated by each subset of connections (the respective
q-arrow or informational relationship). The tent literally
takes shape when the connections are engaged and specify
actual repertoires. Perhaps an even more daring metaphor
would be the following: whenever the mechanisms of a
complex unfold and specify informational relationships, theflower of experience blooms.
From phenomenology to geometry
The notions just sketched aim at providing a framework
for translating the seemingly ineffable qualitative properties
of phenomenology into the language of mathematics, spe-
cifically, the language of informational relationships (q-
arrows) in Q. Ideally, when sufficiently developed, such
language should permit the geometric characterization of
phenomenological properties generated by the human brain.
In principle, it should also allow us to characterize the
phenomenology of other systems. After all, in this frame-
work the experience of a bat echo-locating in a cave is just
another shape in Q and, at least in principle, shapes can be
compared objectively.
At present, due to the combinatorial problems posed by
deriving the shape of the quale produced by systems of just
a few elements, and to the additional difficulties posed by
representing such high-dimensional objects, the best one
can hope for is to show that the language of Q can capture,
in principle, some of the basic distinctions that can be made
in our own phenomenology, as well as some key neuropsy-
229CONSCIOUSNESS AS INTEGRATED INFORMATION
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
15/27
chological observations (Balduzzi and Tononi, unpubl.). A
short list includes the following:
(i) Experience is divided into modalities, like the classic
senses of sight, hearing, touch, smell, and taste (and several
others), as well as submodalities, like visual color and visual
shape. What do these broad distinctions correspond to in Q?
According to the IIT, modalities are sets of densely entan-
gled q-arrows (modes) that form distinct sub-shapes in the
quale; submodalities are subsets of even more densely en-
tangled q-arrows (sub-modes) within a larger mode, thus
forming distinct sub-sub-shapes (Fig. 8). As a two-dimen-
sional analog, imagine a given multimodal experience as the
shape of the three-continent complex constituted by Europe,
Asia, and Africa. The three continents are distinct sub-
shapes, yet they are all part of the same landmass, just as
modalities are parts of the same consciousness. Moreover,
within each continent there are peninsulas (sub-sub-shapes),
like Italy in Europe, just as there are submodalities within
modalities.
(ii) Some experiences appear to be elementary, in that
they cannot be further decomposed. A typical example is
what philosophers call a quale in the narrow sensesay a
pure color like red, or a pain, or an itch: it is difficult, if not
impossible, to identify any further phenomenological struc-
ture within the experience of red. According to the IIT, such
elementary experiences correspond to sub-modes that do
not contain any more densely entangled sub-sub-modes
(elementary modes, Fig. 8).
C
A
=2 bits
MIP
1
1
1
1
0001
0010
0011
0100
0101
0110
0111
1010
10111
101
1001
110
011
10
0000
10
00
COPY
MIP{c
12,c
34}
{c12
}
{c34
}
{ }
COPY
1
2
3
4
B
1
1
1
1
0001
0010
0011
0100
0101
0110
0111
1000
1001
10101
011
1100
1110
11/21/1600000
1101
MIP1
2
3
4
D
Figure 7. The tent analogy. (A) The system of Fig. 2A / Fig. 5. (B) The q-edges converging on the
minimum information partition of the system (MIP) form the natural base on which the complex rests, depictedas a tent. The informational relationships among the parts are built on top of the informational relationships
generated independentlywithinthe minimal parts. From this perspective the q-arrow (in black) is simply the
tent pole holding the quale up above its base; the length (divergence) of the pole expresses the breathing room
in the system. The thick gray q-arrow represents the information generated by the entire system. (C) The system
of Fig. 2A. The quale (not) generated by the two photodiodes considered as a single system. As shown in Fig.
2A, the system reduces to two independent parts, so it does not exist as a single entity. (D) Note that in this case
the quale reduces to the MIP: the tent collapses onto its base, so there is no breathing room for informational
relationships within the system. The quale generated by each part considered in isolation does exist, corre-
sponding to an identical q-arrow for each couple.
230 G. TONONI
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
16/27
(iii) Some experiences are homogeneous and others are
composite: for example, a full-field experience of blue, as
when watching a cloudless sky, compared to that of a busymarket street. In Q, homogeneous experiences translate to a
single homogeneous shape, and composite ones into a com-
posite shape with many distinguishable sub-shapes (modes
and sub-modes).
(iv) Some experiences are hierarchically organized. Take
seeing a face: we see at once that as a whole it is some-
bodys face, but we also see that it has parts such as hair,
eyes, nose, and mouth, and that those are made in turn of
specifically oriented segments. The subjective experience is
constructed from informational relationships (q-arrows) that
are entangled (not reducible to a product of independent
components) across hierarchical levels. For example, infor-
mational relationships constituting face would be more
densely tangled than unnatural combinations such as seen in
certain Cubist paintings. The sub-shape of the quale corre-
sponding to the experience of seeing a face is then an
overlapping hierarchy of tangled q-arrows, embodying re-
lationships within and across levels.
(v) We recognize intuitively that the way we perceive
taste, smell, and maybe color, is organized phenomenolog-
ically in a categorical manner, quite different from, say,
the topographical manner in which we perceive space in
vision, audition, or touch. According to the IIT, these hard-
to-articulate phenomenological differences correspond to
different basic sub-shapes in Q, such as 2n-dimensional
grid-like structures and pyramid-like structures, which
emerge naturally from the underlying neuroanatomy.
(vi) Some experiences are more alike than others. Blue is
certainly different from red (and irreducible to red), but
clearly it seems even more different from middle C on the
oboe. In the IIT framework, in Q colors correspond to
different sub-shapes of the same kind (say pyramids point-
ing in different directions) and sounds to very different
sub-shapes (say tetrahedra). In principle, such subjective
similarities and differences can be investigated by employ-
ing objective measures ofsimilarity between shapes (e.g.,
considering the number and kinds of symmetries involved
in specifying shapes that are generated in Q by different
neuroanatomical circuits).
(vii) Experiences can be refined through learning and
changes in connectivity. Suppose one learns to distinguish
wine from water, then red wines from whites, then differentvarietals. Presumably, underlying this phenomenological
refinement is a neurobiological refinement: neurons that
initially were connected indiscriminately to the same affer-
ents become more specialized and split into sub-groups with
partially segregated afferents. This process has a straight-
forward equivalent in Q: the single q-arrow generated ini-
tially by those afferents splits into two or more q-arrows
pointing in different directions, and the overall sub-shape of
the quale is correspondingly refined.
(viii) Qualia in the narrow sense (elementary modes)
exist at the top of experience and not at its bottom.
Consider the experience of seeing a pure color, such as red.The evidence suggests that the neural correlate (Crick and
Koch, 2003) of color, including red, is probably a set of
neurons and connections in the fusiform gyrus, maybe in
area V8 (ideally, neurons in this area are activated whenever
a subject sees red and not otherwise, if stimulated trigger the
experience of red, and if lesioned abolish the capacity to see
red). Certain achromatopsic subjects with dysfunctions in
this general area seem to lack the feeling of what it is like
to see color, its coloredness, including the redness of
red. They cannot experience, imagine, remember, or even
dream of color, though they may talk about it, just as we
could talk about echolocation, from a third-person perspec-
tive (van Zandvoort et al., 2007). Contrast such subjects,
who are otherwise perfectly conscious, with vegetative pa-
tients, who are for all intents and purposes unconscious.
Some of these patients may show behavioral and neuro-
physiological evidence for residual function in an isolated
brain area (Posner and Plum, 2007). Yet it seems highly
unlikely that a vegetative patient with residual activity ex-
clusively in V8 should enjoy the vivid perceptions of color
just as we do, while being otherwise unconscious.
The IIT provides a straightforward account for this dif-
ference. To see how, consider again Figure 6A: call r the
Red
Color
Form
Sight
Quale
Sound
Figure 8. Modes. Schematic depiction of modes and sub-modes. A
mode, indicated by a polygon within the quale (light gray with black
border), is a set of q-arrows that are more densely entangled than surround-
ing q-arrows, and can be considered as clusters of informational relation-
ships constituting distinctive sub-shapes in Q. Two different modes
could correspond, for example, to the modalities of sight and sound. A
sub-mode within a mode is a set of q-arrows that is even more densely
entangled (a sub-sub-shape in Q). Color and form could correspond to two
sub-modes within the visual mode. The thin black polygon represents an
elementary mode, which does not contain more densely entangled q-arrows.
Elementary modes could correspond to experiential qualities that cannot be
further decomposed, such as the color red (qualia in the narrow sense.)
231CONSCIOUSNESS AS INTEGRATED INFORMATION
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
17/27
connections targeting the red neurons in V8 that confer
them their selectivity, and non-r (r) all the other connec-
tions within the main corticothalamic complex. Adding r in
isolation at the bottom of Q (null context) yields a small
q-arrow (called the down-set of redor 2r) that points in adirection representing how r by itself shapes the maximum
entropy distribution into an actual repertoire. Schematically,
this situation resembles that of a vegetative patient with V8
and its afferents intact but the rest of the corticothalamic
system destroyed. The shape of the experience or quale
reduces to this q-arrow, so its quantity is minimal ( for this
q-arrow is obviously low) and its quality minimally speci-
fied: as we have seen with the photodiode, r by itself cannot
specify whether the experience is a color rather than some-
thing else such as a shape, whether it is visual or not,
sensory or not, and so on.
By contrast, subtract r from the set of all connections, so
one is left with r. This lesion collapses the q-fold spec-
ified by r in all contexts, including the q-arrow, called theup-set of non-red(1r), which starts from the full contextprovided by all other connections r and reaches the top of
the quale.10 This q-arrow will typically be much longer and
point in a different direction than the q-arrow generated by
r at the bottom of the quale. This is because, the fuller the
context, the more r can shape the actual repertoire. Sche-
matically, removing r from the top resembles the situation
of an achromatopsic patient with a selective lesion of V8:
the bulk of the experience or quale remains intact ( re-
mains high), but a noticeable feature of its shape collapses
(the upset of non-red). According to the IIT, the feature of
the shape of the quale specified by the upset of non-redcaptures the very quality or redness of red.11
It is worth remarking that the last example also shows
why specific qualities of consciousness, such as the red-
ness of red, while generated by a local mechanism, cannot
be reduced to it. If an achromatopsic subject without the r
connections lacks precisely the redness of red, whereas a
vegetative patient with just the r connections is essentially
unconscious, then the redness of red cannot map directly to
the mechanism implemented by the r connections. How-
ever, the redness of red can map nicely onto the informa-
tional relationships specified by r, as these change dramat-
ically between the null context (vegetative patient) and the
full context (achromatopsic subject).
A Provisional Manifesto
To recapitulate, the IIT claims that the quantity of con-
sciousness is given by the integrated information () gen-
erated by a complex of interacting elements, and its quality
by the shape in Q specified by their informational relation-
ships. As I have tried to indicate here, this theoretical
framework can account for basic neurobiological and neu-
ropsychological observations. Moreover, the same frame-
work can be extended to begin translating phenomenology
into the language of mathematics.
At present, the very notion of a theoretical approach to
consciousness may appear far-fetched, yet the nature of the
problems posed by a science of consciousness requires a
combination of experiment and theory: one could say that
theories without experiments are lame, but experiments
without theories are blind. For instance, only a theoretical
framework can go beyond a provisional list of candidate
mechanisms or brain areas and provide a principled expla-
nation of why they may be relevant. Also, only a theory can
account, in a coherent manner, for key but puzzling facts
about consciousness and the brain, such as the association of
consciousness with the corticothalamic but not the cerebel-
lar system, the unconscious functioning of many cortico-
subcortical circuits, or the fading of consciousness during
certain stages of sleep or epilepsy.
A theory should also generate relevant corollaries. For
example, the IIT predicts that consciousness depends exclu-sively on the ability of a system to generate integrated
information: whether or not the system is interacting with
the environment on the sensory and motor side, it deploys
language, capacity for reflection, attention, episodic mem-
ory, a sense of space, of the body, and of the self. These are
obviously important functions of complex brains and help
shape its connectivity. Nevertheless, contrary to some com-
mon intuitions, but consistent with the overall neurological
evidence, none of these functions seems absolutely neces-
sary for the generation of consciousness here and now
(Tononi and Laureys, 2008).
Finally, a theory should be able to help in difficult casesthat challenge our intuition or our standard ways to assess
consciousness. For instance, the IIT says that the presence
and extent of consciousness can be determined, in principle,
also in cases in which we have no verbal report, such as
infants or animals, or in neurological conditions such as
minimally conscious states, akinetic mutism, psychomotor
seizures, and sleepwalking. In practice, of course, measur-
ing accurately in such systems will not be easy, but
approximations and informed estimates are certainly con-
ceivable. Whether these and other predictions turn out to be
compatible with future clinical and experimental evidence,
a coherent theoretical framework should at least help to
systematize a number of neuropsychological and neurobio-
logical results that might otherwise seem disparate (Albuset
al., 2007).
In the remaining part of this article, I briefly consider
some implications of the IIT for the place of experience in
our view of the world.
Consciousness as a fundamental property
According to the IIT, consciousness is one and the same
thing as integrated information. This identity, which is
232 G. TONONI
8/13/2019 conciousness as integrated information - Giulio Tononi.pdf
18/27
predicated on the phenomenological thought experiments at
the origin of the IIT, has ontological consequences. Con-
sciousness exists beyond any doubt (indeed, it is the only
thing whose existence is beyond doubt). If consciousness is
integrated information, then integrated information exists.
Moreover, according to the IIT, it exists as a fundamental
quantityas fundamental as mass, charge, or energy. As
long as there is a functional mechanism in a certain state, it
must existipso factoas integrated information; specifically,
it exists as an experience of a certain quality (the shape of
the quale it generates) and quantity (its height ).12
If one accepts these premises, a useful way of thinking
about consciousness as a fundamental property is as fol-
lows. We are by now used to considering the universe as a
vast empty space that contains enormous conglomerations
of mass, charge, and energygiant bright entities (where
brightness reflects energy or mass) from planets to stars to
galaxies. In this view (that is, in terms of mass, charge, or
energy), each of us constitutes an extremely small, dimportion of what existsindeed, hardly more than a speck of
dust.
However, if consciousness (i.e., integrated information)
exists as a fundamental property, an equally valid view of
the universe is this: a vast empty space that contains mostly
nothing, and occasionally just specks of integrated informa-
tion ()mere dust, indeedeven there where the mass-
charge energy perspective reveals huge conglomerates. On
the other hand, one small corner of the known universe
contains a remarkable concentration of extremely bright
entities (where brightness reflects high ), orders of mag-
nitude brighter than anything around them. Each bright-star is the main complex of an individual human being
(and most likely, of individual animals).13 I argue that such
-centric view i